Rather than training another AI network to detect fakes, I think we need to use semantic likelihood. That is, what is the probability of this picture being real

Rather than training another AI network to detect fakes, I think we need to use semantic likelihood. That is, what is the probability of this picture being real based on its content? Low probabilities demand higher proof. I think this is what my brain does.